HIGHER EDUCATION AND GENERATIVE AI (GENAI) IN ISRAEL:

2.1 Artificial Intelligence (AI)

Artificial Intelligence (AI) is a branch of Sciences/Computer Science that focuses on creating intelligent machines that can perform tasks that typically require human intelligence, such as visual perception, speech & text processing, decision-making support, and dialog conversation analysis and generation. AI systems are designed to learn from data, adapt to new situations, and improve their performance over time with various kinds of machine learning algorithms.

In education and learning systems, AI has the potential to revolutionize the way we teach and learn. AI-powered systems can personalize learning experiences for individual students, and provide real-time feedback and support. AI can also help teachers and administrators to identify areas where students are struggling and provide targeted interventions to help them overcome and succeed. Following are two examples of AI-powered education and learning systems:

• Intelligent Tutoring Systems: These systems use machine learning algorithms to personalize learning experiences for individual students, providing them with feedback and support based on their performance. DeamBoxLearning is an example of such a system.
• Adaptive Learning Platforms enabling independent learning: These platforms use machine learning algorithms to adapt to the learning preferences of individual students, providing them with personalized content. Khan Academy and its advanced version Khanmigo is an example of such a platform.

Artificial Intelligence is usually classified into one of four types: General AI, Narrow AI, Edge AI and Generative AI our focus in this chapter.

2.2 Generative AI (GenAI)

Generative AI is a subfield of Artificial Intelligence (AI) that focuses on creating machines that can generate new content, such as images, music, and text. Unlike other AI systems that are designed to recognize patterns and make predictions based on existing data, Generative AI systems are designed to create new data that has never been seen before and is remarkably on the high level and quality of human creativity.

The history of GenAI systems can be traced back to the 1950s. However, it wasn't until the 1990’s that GenAI began to gain a status as a field of study. In 1997, a program called AARON was developed by artist Harold Cohen, which could generate original artwork using a set of rules and algorithms. In the early 2000’s, GenAI began to evolve rapidly, thanks to advancements in machine learning and deep learning. In 2014, a team of researchers at Google developed a GenAI system called DeepDream, which could generate images by analyzing existing images and identifying patterns and features. In 2016, a GenAI system called WaveNet was developed by researchers at Google, which could generate realistic-sounding speech by modeling the human voice. This system was a significant breakthrough in the field of GenI and paved the way for the development of other systems.

In 2018, a Generative AI system called GPT2 wad developed by OpenAi , which could generate human-like text by analyzing large amounts of data. This system was capable of generating coherent and convincing text, which raised concerns about the potential misuse of GenAI. These capabilities has continued to evolve, with GPT 3,4 and new systems that are being developed by Google, Apple and Amazon generating more complex and sophisticated content. GenAI has the potential to transform many industries and change the way we interact with machines. As technology continues to advance, a whole echo-system has evolved, and we can expect to see even more exciting developments in the field of Gen AI in the years to come.

2.2.1 NLP

Natural Language Processing (NLP) is a subfield of Artificial Intelligence (AI) that deals with the interaction between computers and human language (Text/Speech). Over the past 30 years, NLP has undergone significant changes, and its history and timeline are important for the understanding of AI development in general and Generative AI specifically.

In the early 1990s, NLP was primarily focused on rule-based systems, where experts would manually create rules to process language. However, this approach was limited in its ability to handle the complexity and variability of natural language. In the late 1990s, statistical methods were introduced, which allowed computers to learn from large amounts of data. This approach led to significant improvements in NLP, particularly in tasks such as language translation and speech recognition. In the early 2000s, machine learning techniques such as deep learning and neural networks were introduced, which further improved the accuracy of NLP systems. These techniques allowed computers to learn from vast amounts of data and make predictions based on that data. As a result, NLP systems became more sophisticated and could handle more complex tasks such as sentiment analysis and natural language understanding. In recent years, there has been a significant shift towards the use of pre-trained language models such as BERT and GPT3. These models are trained on massive amounts of data and can perform a wide range of NLP tasks with high accuracy. They have revolutionized the field of NLP and have made it possible to develop applications such as chatbots and virtual assistants that can understand and respond to in a conversational dialog.

NLP has evolved to become a sophisticated field that can handle complex tasks such as natural language understanding and Sentiment Analysis. One of the most challenging NLP tasks is ambiguity resolution, (disambiguation)resulting from the high level of language ambiguity at all levels; lexical, semantics, pragmatics and referential. Even highly pre-trained statistical models are challenged handling cases easily resolved by humans especially in pragmatic context.

2.2.2 Transformers and LLMs

Large Language Models (LLMs) are a type of machine learning model that is trained on large amounts of text data. The goal of an LM is to learn the patterns and structures of language so that it can generate new text that is similar to the training data. LLMs are used in many GeneAI systems, including text generators and chatbots. The problem and challenge of LLMS is that it’s based on the train data which is harvested automatically from the web sources. If these data contain fake or biased information it will be hard to detect and final results will be tented. That is why many of the systems use “a human in the loop” methodology to verify the process and the quality of the data and results.

Transformers (Transformers) are a type of deep learning model that was introduced in 2017. They are designed to process sequential data, such as text or speech, and are particularly effective at generating natural language. In the Transformer architecture, the input sequence is first transformed into a set of vectors called embeddings. These embeddings are then processed by a series of layers that perform operations such as attention mechanism and feedforward neural networks. The output of the final layer is then transformed back into the original sequence. Transformers use a technique called self-attention. This mechanism in the transformer works by computing a set of attention weights for each position in the input sequence. These attention weights indicate how much attention should be paid to each position when generating a more accurate and coherent output.

Two more mechanisms are widely used in Generative AI and are briefly mentioned here: Generative Adversarial Networks (GANs). GANs are a type of deep learning model that consists of two neural networks: a generator and a discriminator. Recurrent Neural Networks (RNNs) designed to process sequential data and are particularly effective at generating text and speech. They work by maintaining a hidden state that captures the context of the input sequence and using it to generate the output in case of ambiguity for example. 

2.3 Generative AI - why now?

The main use of AI as it developed since the 1990s was mainly in research laboratories and academic institutions and very little by real mainstream users. Since the end of 2022, we have witnessed a huge increase in awareness, familiarity and use of AI tools, especially those considered GenAI: ChatGPT, Midjourney, DAL-E-2, DID and more, among users without prior knowledge or training in AI.

The question that arises is “why now?” We offer four main reasons:

1. Technology is mature, fast and relatively reliable: Core technologies has reached a pick of evolvement both in terms of the computing infrastructure, cloud computing and processing power but also in the availability of strong algorithms, data harvesting and storage. These enable a fast and mature handling of complex real time AI models.
2. User interface is simple & freely available: The use of simple language interfaces, natural and intuitive interaction have removed the barrier for the naïve users. As will be discussed below prompt engineering is crucial but the set of available tools and services around the AI engines such as GPT is growing at the moment and enables almost perfecting any task user desires.
3. Users’ readiness to explore and adapt: As a result of the high, partly free, and simple availability, in conjunction with the technology savvy student’s population, and the effect of social networks and globalization, the readiness of individual users to explore AI application is great. The barriers of technological fear were removed, and the added value of ease of use created motivation among users from a very early age.
4. Search engine - Google just isn’t enough: New user’s (and young) behavior and needs in educational processes is leading to the demand of students to get more than just information in a list of links. In light of such accessibility to best practices and organized ready-made knowledge, the wish is to get task and action done and not just researching for data source. While Google is a search engine that provides information, GenAI applications analyze the information, draw conclusions, and provide intelligent answers to search queries.

In addition to research, higher education systems in the world and in Israel are responsible for teaching and learning, that is, for the transfer and acquisition of knowledge. The imparting of knowledge is the responsibility of the faculty. On the part of the students, learning is a process of acquiring knowledge, the results of which will lead to a change in professional abilities in the present and/or in the future. The common teaching and learning method in the academy are based on a physical location (campus, lecture hall, classroom), a limited learning time frame (year, semester, day, hour) a pre-structured curriculum (degree, specialization, course, workshop, exercise), known scope (semester hours, credit points), assessment of learning outcomes known in advance (exam, work) and a distinctive organizational structure (faculties, departments).

Until the COVID-19 crisis, at the beginning of 2020, most of the teaching and learning in the Israeli academy (except for the Open University of Israel, which specializes in distance learning) took place on the institution campus. In the common systemic model, learning was seen as a mechanical process and the teaching approach as industrial in the format of a mass, centralized, uniform and standardized production line. In such a system, the emphasis is on the evaluation of the final product (test, work) and its measurable dimensions (a grade) and less on the process and the personal learning experience of the students.

The digital revolution and the opening of the Internet to the public, starting at the end of the nineties of the last century, essentially did not change the classical teaching and learning processes. Although extensive and huge information sources that can be reached with a click have been added, the industrial concept remains the same.

Another development of the digital revolution is the opening of course websites that accompany the classroom teaching and are supervised by the teaching staff. The purpose of such a learning site (usually within a learning management system such as Moodle) is to be a content anchor (such as lesson presentations, remote access to databases), interactive (forum, chat) and learning management. That is, these sites enriched the educational process but did not essentially change the concept of the industrial teaching method.

The COVID-19 crisis forced an immediate route recalculation. The transition to emergency and distance learning was similar to a hurricane shaking the normal, familiar and well-known in the academy. Higher education institutions all over the world had to adapt to the changing reality, which did not allow it to continue operating in the accepted format until now. Not surprisingly, the lecturers, most of whom did not have knowledge or were not trained to teach in this format at all, expressed distress at the complexity involved in interacting with the students in remote teaching and the technological challenges². A similar distress was conveyed by students in a survey conducted by the Edmund de Rothschild Foundation where they expressed their difficulties in continuing academic studies in a format of distance learning³..

4.1 Introduction

In this section we will present key implications of integrating GenAI into learning and teaching in higher education in Israel. As we are in the midst of a "tornado, this incubation phase allows us to present a partial answer, one that is current and valid at the time of writing the position paper (October 2023). We have chosen to briefly present key characteristics that we believe will be dramatically affected by the development of GenAI. These key characteristics are divided into three key categories:

4.2 Improving the learning process

4.2.1 ”A buddy to study with" - AI can assist in the learning process by providing information, asking questions, cultivating reflective-critical thinking, solving problems, and deepening the understanding of concepts. It can offer new perspectives and help see things in different and divers ways. Concrete examples are:

• GenAI-as a personal assistance for the student learning process while the lecturer humanly, emotionally, socially and personally supports the student.
• Social learning in groups where GenAI tool plays the role of a team member who participates and aid the conversation.
• Creating simulations through role-playing with the GenAI as a partner and/or guide.

Expanding the ways of learning can empower the students and encourage them to take responsibility for their personal learning process as well as to develop self-proactive learning skills instead of passively absorbing knowledge (as is customary in lecture rooms). However, over-reliance has the potential to harm intellectual development as it may cause the student to transfer the responsibility for learning to GenAI without engaging in critical thinking (or thinking at all).

4.2.2 Personalization in learning - an educational approach according to which a learning program must be adapted personally (student-centered learning), in real time and continuously to each student according to his needs - learning rate, interest, strengths, etc⁸. With the help of AI tools, the learning process can be adapted to the individual need and provide customized feedback and guidance. This can help promote an interactive learning experience that contributes to the student's interest and engagement and allows the student to understand the learning process qualitatively and according to his/her personal characteristics. However, over-reliance on technology may lead the student feel detachment from the human lecturer and as a result develop feelings of alienation, "getting lost", and a decrease in motivation to learn.

4.2.3 Shortening time in teaching and learning - GenAI tools can enable an immediate response to questions or requests for assistance from a lecturer and student. This can help reduce workload for teaching staff. For example: relying on the activation of automatic evaluation and review mechanisms that provide immediate feedback to students, and for students the option of facilitation of homework and papers preparation. For example: using GenAI tools to find sources, initial drafting of work, finding keying typos and more. These easier paths can help learners focus on more complex or challenging aspects of the teaching-learning process. However, over-reliance on shortcuts and over-reliance on GenAI tools, in particular for writing assignments and exercises, may lead to superficial products that indicate a lack of understanding.

4.3 Learning Skills

4.3.1 Critical Thinking - is the ability to think clearly and rationally, understanding the logical connection between ideas. It's about being active (not reactive) in the learning processes, and it involves being open-minded, inquisitive, and able to think in a reasoned way. Critical thinking is also, a balanced and reflective thinking that focuses on deciding ‘what to believe and what to do’⁹. This skill is considered a solid learning skill as it focuses on "how to think" instead of "what to think". Solid skills like critical thinking are more immune to GenAI advances because they represent many of the tasks that GenAI systems struggle with today. This is in contrast to temporal skills, such as a programming language or a specific tool, which lose value quickly due to technological progress¹⁰.

In our opinion, the skill of critical thinking should be put at the top of academic discourse. In an era where texts will be created automatically, human contribution will be in examining the content, criticizing it and amending it. The variety of sources, and points of view allows for a critical and comparative examination of the content. Avoiding critical thinking and over-reliance on AI tools can create for the learner a false sense of security in the precision of the created content. This may lead students to trust the results generated by AI without questioning their validity or accuracy.

4.3.2 Creativity – is the process of producing original and relevant ideas for the given situation. Creative thinking consists of originality and flexibility¹¹.. It can be an inspiration for new and creative ideas, suggest new practices, provide new and original perspectives on an educational issue. However, over-reliance on GenAI tools may limit personal creativity and impair independent human voice.

4.3.3 Originality - refers to creating or inventing new learning products. Original work is not work that is received from others or work that is copied or based on another's work, such as GenAI. The challenge in training learners to prepare learning materials and products that they themselves have created, is finding the balance between using GenAI as a "scaffolding" for learning process and just transferring the responsibility for preparing the papers and exercises to GenAI tools.

4.4 Ethics & Ethical Aspects

4.4.1 Ethics - is a branch of philosophy that examines moral principles and values to determine what is right and wrong in different situations. In the academic context, ethical behavior is characterized by integrity and respect for rights and intellectual property. The ethical complexity of integrating GenAI tools invites possibilities for developing critical thinking, establishing an ethical code and creating a safe and open teaching-learning environment that can help develop an ethical compass among students and faculty alike. As a result of the development of GenAI technology, it may become difficult to evaluate works and content submitted by students and faculty members, without knowing whether they were written by a human or a machine.

Moreover, it will be difficult to trace the knowledge sources and its lack of transparency, which may harm the privacy and protection of the intellectual property of the content owners, and even harm the privacy and security of the users themselves. Relying on existing content also raises the concern of creating bias in the new content that will be generated, which will specifically exclude disadvantaged populations (because the written text is based on what the "majority" has written on the subject so far). Added to this is the fear of "hallucinations”, in cases machines sometimes invents content. Thus, require careful human caution to identify these cases. Since in the era of GenAI it is difficult to detect unethical conduct and even more so, to enforce it through disciplinary means, a difficulty may raise in balancing between academic freedom and personal responsibility.

4.4.2 Algorithmic bias - as a private case of ethics according to which GenAI technologies are based on collecting data from users’ activity according to an algorithmic basis unknown to the end users ("black box") and understanding the automatic decisions that these systems produce.

Autonomous systems may increase or perpetuate existing biases and thus increase the deepening of discrimination and bias against disadvantaged population

4.4.3 Data protection and privacy – as a private case of ethics, it is known that GenAI technologies are based on the collection and analysis of personal and group information of learners. There is a real risk of information leakage to unauthorized sources/factors. The challenge is in preventing information leakage through technological means as well as establishing a clear protection strategy and privacy policy for every individual in the society.

4.4.4 Accessibility to all (?) - Availability and accessibility of GenAI technologies to every person regardless of socio-demographic status or any other irrelevant characteristic invites the expression for democratization of human knowledge since, GenAI tools such as ChatGPT include all human knowledge accumulated on the Internet (with certain reservations). It follows that GenAI technologies can make teaching and learning open to every student regardless of disabilities or limitations, thus promoting a culture of inclusion and diversity (DIEB - Diversity, Inclusion, Equality & Belonging) that emphasizes differences between learners, and develops a sense of belonging. The richness of the content, the exposure to a variety of opinions and points of view, can encourage student awareness of the complexity of acquiring knowledge and developing empathy and tolerance for different opinions. However, the lack of clarity regarding the costs of the technology and the accessibility it will provide, raises a fear of harming equality, diversity and inclusion and may actually increase disparities instead of reducing them.

Finally, we have put together key recommendations for decision-makers, including academic staff, preparing for the era of GenAI academic learning and teaching:

1. Awareness of the disruptive change - this is not hype! - Some believe that humanity is currently facing the beginning of the greatest technological disruption in history. Either way, academic institution managements and faculty must become deeply familiar with the revolutionary possibilities that GenAI applications offer for teaching, learning and research. This awareness should lead to the development of an institutional strategy. That is, formulating an institutional policy for the adoption of GenAI applications and an adapted design of the study programs, including the required allocation of resources.
2. Maintaining the relevance of the academy - For years arguments have been heard about the irrelevance of the academy as the responsible and trusted body for the training future generations of workers. Academia, as a research body that has also undertaken the process of training professionals in a variety of fields, is required to constantly update the area of expertise as well as its learning methods¹³. According to most estimates, the penetration of GenAI will have a dramatic effect on the global job market and on professions that require academic training. For the academy, the challenge of relevancy becomes more substantial as there is the potential for an even greater gap between the nature of the students' training and the work environment, they will encounter upon transitioning to the labor market. Accordingly, academia must already take steps today to understand the nature of the change in each content area and reflect it in teaching and assessment processes in the study framework.
3. Applied research is needed! - Research lies at the heart of academic endeavors. The advent of GenAI should not diminish the importance of continued research in this domain. However, in the short research time window in which we are now, a focus on conducting applied research is required. That is, research that examines the execution and actual implementation of GenAI-powered learning and teaching processes and compares them to traditional learning-teaching processes. The findings of these studies will help in understanding the contribution of the technologies for planning a responsible change process in the design of the learning and teaching processes.
4. Changing teaching and assessment practices - In many higher education institutions, lectures remain the predominant teaching method. This approach involves imparting knowledge to a large audience of students, who are often passive recipients expected to assimilate the information. Despite being a staple for centuries, the efficacy of this method is limited, especially when compared to strategies that foster active learning and stimulate higher-order thinking¹⁴. One notable example of an active learning approach is Project-Based Learning (PBL). PBL positions the student at the heart of the learning journey. It encompasses real-world problem-solving, data analysis, decision-making, collaboration with peers, and the creation of a tangible learning outcome, all while chronicling the learning trajectory ¹⁵. Similarly, when it comes to assessment, many institutions still rely heavily on traditional tests that primarily gauge a student's memorization skills, often sidelining the practical application of acquired knowledge. It's imperative to align assessment methods with learning objectives and to diversify evaluation techniques. This promotes not just rote learning but also meaningful comprehension and critical thinking. It is worth noting that these recommended shifts in teaching and assessment are not solely prompted by the advent of GenAI. However, the emergence of this era certainly underscores the urgency of these changes.
5. Imparting learning literacies adapted to the GenAI era - Students should be equipped with AI literacy, encompassing the knowledge, skills, and competencies needed for informed and discerning navigation within the AI ecosystem. Faculty must teach the students about how the AI applications work and at the same time instruct them to use these applications only after acquiring thinking scaffolding in the content area on which they are looking for information. By adopting this approach, students are better positioned to assess the quality of the responses they obtain from AI tools. Such scrutiny fosters the development of vital skills in an era where AI is deeply woven into the fabric of society. These essential skills encompass critical thinking, advanced cognitive abilities, creativity, and originality. Moreover, it's a misconception to presume that students inherently grasp the intricate ethical dilemmas tied to GenAI applications. As such, ethical reasoning and academic integrity should be seamlessly integrated into the curriculum.
6. Training faculty - The teaching staff holds the responsibility for integrating GenAI within academic learning. Consequently, it's imperative to bolster the faculty's familiarity with these new tools, understanding both their potential and limitations. Faculty members should be motivated to engage hands-on, broadening and deepening their expertise in the domain. Methods to train the faculty can encompass workshops, discussion groups, one-on-one training sessions, the dissemination of guidelines, sharing of successful experiences, and more.